OpenAI Text Watermark Detection Drops to 17% After Minor Edits
· 3h ago · Originally reported by PPC Land
OpenAI's text watermark detection catches about 80% of AI-generated 200-token passages at a 1% false-positive rate, but effectiveness drops to 17% when just 25% of words are changed. API users can opt in now, and ChatGPT and Codex outputs in the EU will be automatically marked within weeks.
Why it matters
AI-generated content fuels low-quality MFA sites that drain advertiser budgets in programmatic. Reliable watermarking helps buyers verify content authenticity, but easy circumvention through minor edits limits its practical value.
Explain it like I’m new here
It's like a secret invisible stamp on AI-written text that fades away if you just rearrange a few words.
Related stories
Integrate Launches MCP Connector to Bring B2B Marketing Data Into AI Workflows
Integrate has released Integrate MCP, a connector that lets B2B marketers query campaign, lead, spend, pipeline and ABM data through AI tools while preserving existing permissions and approval controls. The product aims to let teams act on…
Razorpay Engage Enables Indian Brands to Run Ad Campaigns on ChatGPT
Razorpay has expanded its Engage suite to let Indian merchants launch and manage ad campaigns directly on ChatGPT. The rollout builds on the fintech firm's existing collaboration with OpenAI focused on AI-powered commerce.
Amazon DSP Centres AI Agents Across Buying, Creative and Targeting
Amazon's demand-side platform is integrating AI agents into core advertising workflows including media buying, creative generation and audience targeting. The approach, highlighted by Ankur Rathi, signals deeper automation across the…
ChatGPT's Reported Advertising Push Raises Questions for Consumers and Advertisers
OpenAI appears to be moving toward introducing advertising within ChatGPT, according to an analysis piece examining the implications. The article explores what this shift could mean for users, the advertising industry, and OpenAI's own…